Parser stack corruption: error during backtracking treated incorrectly like YYERROR

[teshields]

I experienced a strange memory corruption problem during backtracking that I believe is due to an error in btyaccpa.ske:

• All parse algorithm errors are treated as if the YYERROR macro has been called.
• The behavior is then to first pop the rhs of the "rule" internally generated for the action containing YYERROR.
• If the error was not caused by the YYERROR macro, this mangles the stack.

I have a patch (attached) to btyaccpa.ske that avoids the problem, although I haven't dug into the skeleton code sufficiently to understand if the patch breaks something else.

I have also attached zip file containing an example that demonstrates the problem (the Makefile expects 'btyacc' and 'btyacc.bug' as 2 separate executables, built with and without the patch).

btyacc-20170602.btyaccpa.ske.patch.txt

btyacc-bug.zip

[teshields]

I forgot to note that the zip file includes the file 'test-input' that results in the parse stack corruption when the 'gc' executable is built with the distribution btyacc; with the patched btyacc, it parses with no problems.

[teshields]

I went back and looked at the version of btyaccpa.ske in the btyacc 3.0 distribution still available on the Siber Systems web site.

In the location of my patch, that old version of btyaccpa.ske is the same as my patch.

I haven't searched through the change history for btyaccpa.ske yet, but I assume that you made the change that my patch 'backs out' to address some other problem. So my patch may not be the end solution.

[ChrisDodd]

Fixed by 92963ccfda1eb774b65b666fc9ae9a0184ab1ad3.